KIIT Researchers' Findings on AI-Led Breast Cancer Detection Published in Prestigious Lancet Journal

KIIT Researchers' Findings on AI-Led Breast Cancer Detection Published in Prestigious Lancet Journal

Bhubaneswar, July 2: A publication authored by researchers from KIIT-DU, in the prestigious medical journal The Lancet, has highlighted the importance of considering age-related tumour biology when evaluating artificial intelligence (AI)-based breast cancer screening.The KIIT researchers argue that incorporating these biological differences could strengthen the assessment of AI-assisted mammography programmes.The article, authored by Dr. Pratap Kumar Jena, Associate Professor, Mr. Sabyasachi Shukla, PhD Scholar, and Dr. Ramya Pinnamaneni, Director of the School of Public Health at KIIT, argues that AI-powered mammography tools should not be evaluated solely on technological performance. Instead, they should also consider the biological differences in tumour growth across age groups to generate more reliable evidence on the effectiveness of breast cancer screening.The KIIT researchers have raised the issue while responding to the findings of the Swedish MASAI (Mammography Screening with Artificial Intelligence) trial, one of the world’s largest studies evaluating AI in breast cancer screening. While the trial reported that AI-assisted screening detected more cancers and performed better than conventional double-reading by radiologists, the KIIT team believes that the study did not sufficiently account for age-related differences in tumour biology.According to the researchers, breast cancers do not behave the same way in women of all ages. Tumours in younger women often grow faster, while those in older women tend to grow more slowly and remain detectable for longer periods.This “window of detectability” can significantly influence the apparent performance of AI systems, making them appear either more or less effective than they actually are.“If AI-based screening assessments do not account for age-related differences in tumour growth and the window of detectability, estimates of screening effectiveness may be incomplete,” said Dr. Pratap Kumar Jena. He added that integrating biological factors into AI evaluation frameworks could improve predictive accuracy and reduce errors in assessing AI performance.The publication represents another scholarly contribution from KIIT researchers to the evolving discussion on AI-assisted healthcare. Earlier this year, the same team published a correspondence in The Lancet Digital Health, highlighting methodological considerations in evaluating AI-assisted mammography, including the importance of accounting for the preclinical "window of detectability" (or sojourn time) and clarifying key screening metrics.Building on that, the researchers have recommended that future AI-supported mammography studies incorporate mathematical models that account for age-related variations in cancer growth. Such an approach, they say, would provide a more accurate assessment of AI’s true clinical value and help healthcare systems make better screening decisions for women across different age groups.